livebench/reasoning
收藏资源简介:
LiveBench是一个为大型语言模型(LLMs)设计的基准测试,旨在限制测试集污染并进行客观评估。它具有以下特点:每月发布新问题,问题基于最新发布的数据集、arXiv论文、新闻文章和IMDb电影摘要;每个问题都有可验证的客观真实答案,允许准确自动评分,无需使用LLM评判;目前包含18个不同任务,分布在6个类别中,并计划发布更难的挑战。这是LiveBench的instruction_following类别。
LiveBench is a benchmark for LLMs designed with test set contamination and objective evaluation in mind. It has the following properties: it releases new questions monthly, with questions based on recently-released datasets, arXiv papers, news articles, and IMDb movie synopses; each question has verifiable, objective ground-truth answers, allowing for accurate and automatic scoring without the use of an LLM judge; it currently contains a set of 18 diverse tasks across 6 categories, with plans to release new, harder tasks over time. This is the instruction_following category of LiveBench.
数据集概述
数据集特征
- question_id: 数据类型为字符串。
- category: 数据类型为字符串。
- ground_truth: 数据类型为字符串。
- turns: 数据类型为字符串序列。
- grouping: 数据类型为字符串。
数据集分割
- test:
- 数据量: 195226 字节
- 示例数量: 150
数据集大小
- 下载大小: 61439 字节
- 数据集总大小: 195226 字节




